Downloads · 30 days
0
polt-transformer
peiti/polt-transformer
polt-transformer is a machine learning model from peiti. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
9 specialized PolarisModule experts (14.5M params each, NoctusLayer + byte-level transformer).
Access
Public
Updated May 30, 2026
Repo size
1.5 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt1 GB · 100%
At a glance
- Access
- Public
- Created
- May 30, 2026
- Updated
- May 30, 2026
- SHA
- ee235f3e
Model card
From the Hugging Face model README
PHANTOM PolarisModule — Expert Checkpoints
9 specialized PolarisModule experts (14.5M params each, NoctusLayer + byte-level transformer).
Experts
| File | Domain | Lang | Loss |
|---|---|---|---|
experts/polaris_astro_cz.pt | Astronomy | CS | 2.57 |
experts/polaris_dialog_cz.pt | Conversation | CS | 1.96 |
experts/polaris_news_cz.pt | News | CS | 2.53 |
experts/polaris_legal_cz.pt | Legal | CS | 2.48 |
experts/polaris_jwst_en.pt | JWST/Space | EN | 2.59 |
experts/polaris_science_en.pt | Science | EN | 2.67 |
experts/polaris_creative_cz.pt | Literature | CS | - |
experts/polaris_helix_emo.pt | Emotion (HELIX-9) | CS | - |
experts/polaris_nexus_int.pt | Multilingual | CS+EN | - |
Architecture
- ByteEmbed: byte-level (0–255), d=512, pos=2048
- TransformerEncoder: 4 layers, 8 heads, FFN=2048, GELU, norm_first
- NoctusLayer: bipolar (tanh) + lateral inhibition (Conv1d grouped) + gating
- LM Head: 512→259 logits
MoE
See polaris_moe_v1.pt for the consolidated MoE model (top-2 router, 9 experts).
- Created
- May 30, 2026
- Updated
- May 30, 2026